/* This 12-byte structure is used to improve the accuracy of computation of transcendental functions. Intended to be used to get results better than 8-byte computation allows. 9-byte would probably be sufficient.
*/ typedefstruct { unsignedlong lsw; unsignedlong midw; unsignedlong msw;
} Xsig;
/* Macro to extract the most significant 32 bits from a long long */ #define LL_MSW(x) (((unsignedlong *)&x)[1])
/* Macro to initialize an Xsig struct */ #define MK_XSIG(a,b,c) { c, b, a }
/* Macro to access the 8 ms bytes of an Xsig as a long long */ #define XSIG_LL(x) (*(unsignedlonglong *)&x.midw)
/* Need to run gcc with optimizations on to get these to actually be in-line.
*/
/* Multiply two fixed-point 32 bit numbers, producing a 32 bit result.
The answer is the ms word of the product. */ /* Some versions of gcc make it difficult to stop eax from being clobbered. Merely specifying that it is used doesn't work...
*/ staticinlineunsignedlong mul_32_32(constunsignedlong arg1, constunsignedlong arg2)
{ int retval; asmvolatile ("mull %2; movl %%edx,%%eax":"=a" (retval)
:"0"(arg1), "g"(arg2)
:"dx"); return retval;
}
/* Add the 12 byte Xsig x2 to Xsig dest, with no checks for overflow. */ staticinlinevoid add_Xsig_Xsig(Xsig *dest, const Xsig *x2)
{ asmvolatile ("movl %1,%%edi; movl %2,%%esi;\n" "movl (%%esi),%%eax; addl %%eax,(%%edi);\n" "movl 4(%%esi),%%eax; adcl %%eax,4(%%edi);\n" "movl 8(%%esi),%%eax; adcl %%eax,8(%%edi);\n":"=g"
(*dest):"g"(dest), "g"(x2)
:"ax", "si", "di");
}
/* Add the 12 byte Xsig x2 to Xsig dest, adjust exp if overflow occurs. */ /* Note: the constraints in the asm statement didn't always work properly with gcc 2.5.8. Changing from using edi to using ecx got around the
problem, but keep fingers crossed! */ staticinlinevoid add_two_Xsig(Xsig *dest, const Xsig *x2, longint *exp)
{ asmvolatile ("movl %2,%%ecx; movl %3,%%esi;\n" "movl (%%esi),%%eax; addl %%eax,(%%ecx);\n" "movl 4(%%esi),%%eax; adcl %%eax,4(%%ecx);\n" "movl 8(%%esi),%%eax; adcl %%eax,8(%%ecx);\n" "jnc 0f;\n" "rcrl 8(%%ecx); rcrl 4(%%ecx); rcrl (%%ecx)\n" "movl %4,%%ecx; incl (%%ecx)\n" "movl $1,%%eax; jmp 1f;\n" "0: xorl %%eax,%%eax;\n""1:\n":"=g" (*exp), "=g"(*dest)
:"g"(dest), "g"(x2), "g"(exp)
:"cx", "si", "ax");
}
/* Negate (subtract from 1.0) the 12 byte Xsig */ /* This is faster in a loop on my 386 than using the "neg" instruction. */ staticinlinevoid negate_Xsig(Xsig *x)
{ asmvolatile ("movl %1,%%esi;\n" "xorl %%ecx,%%ecx;\n" "movl %%ecx,%%eax; subl (%%esi),%%eax; movl %%eax,(%%esi);\n" "movl %%ecx,%%eax; sbbl 4(%%esi),%%eax; movl %%eax,4(%%esi);\n" "movl %%ecx,%%eax; sbbl 8(%%esi),%%eax; movl %%eax,8(%%esi);\n":"=g"
(*x):"g"(x):"si", "ax", "cx");
}
#endif/* _POLY_H */
Messung V0.5
¤ Dauer der Verarbeitung: 0.14 Sekunden
(vorverarbeitet)
¤
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.